[
https://issues.apache.org/jira/browse/CASSANDRA-21702?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dmitry Konstantinov updated CASSANDRA-21702:
--------------------------------------------
Fix Version/s: 6.0
7.0
(was: 7.x)
(was: 6.0.x)
Source Control Link:
https://github.com/apache/cassandra/commit/3e939c8bb7507e5a2068df7a4778b54304749d2b
Resolution: Fixed
Status: Resolved (was: Ready to Commit)
> Reduce prepared statement binding overhead for Cassandra stress
> ----------------------------------------------------------------
>
> Key: CASSANDRA-21702
> URL: https://issues.apache.org/jira/browse/CASSANDRA-21702
> Project: Apache Cassandra
> Issue Type: Improvement
> Components: Tool/stress
> Reporter: Dmitry Konstantinov
> Assignee: Dmitry Konstantinov
> Priority: Normal
> Fix For: 6.0, 7.0
>
> Attachments: bind_lookup_overhead.png, stress_bind_after_cpu.html,
> stress_bind_before_cpu.html
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Under high 10-row, 5 value column batch write load Cassandra stress burn a
> lot of CPU to lookup Codecs for bound values: 21.5% of overall CPU
> consumption ("codecFor" to seach in profiles).
> To avoid this overhead we can retrieve required codecs for a prepared
> statement once and then use them explicitly when we bind values to avoid
> codec lookup cost. It drops the codecFor lookup overhead to 0.
> [^stress_bind_before_cpu.html]
> [^stress_bind_after_cpu.html]
> !bind_lookup_overhead.png|width=400!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
